Gemini 3.8 Flash vs Muse Ember
Muse Ember leads by 1.5 capability points (≈95% interval −0.8 to +3.9): statistically tied · P(higher) ≈ 90%
Wins 36 of 57 shared benchmarks (20 lost, 1 tied)
Capability vs. relative cost
Gemini 3.8 FlashMuse EmberClosed weightsOpen weights
Benchmark wins20 – 36 · 57 shared, 1 tied
Comparing matching setups only (1 benchmark excluded) · suites counted once
Area← Gemini 3.8 FlashMuse Ember →
Data: Artificial Analysis, Vals AI, Surge AI and 10 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Ember minus Gemini 3.8 Flash.
| HiL-Bench (Human-in-Loop Benchmark) | Tool & computer use › General agents | 41.5 | 74.0 | +32.5 |
| DAXBench v2 | Software › Data & ML | 87.2 | 62.8 | −24.5 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 19.7 | 40.9 | +21.2 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 4.5 | 24.2 | +19.7 |
| Terminal-Bench 4.0 | Software › Software engineering | 19.7 | 39.4 | +19.7 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1204 | 1526 | +17.7 |
| APEX-SWE | Software › Software engineering | 36.4 | 52.9 | +16.5 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 2.0 | 17.2 | +15.2 |
| HLAB-Vals-1.1 | Professional › Legal | 12.1 | 25.0 | +12.9 |
| Artificial Analysis Coding Agent Index v2.0 | Software › Software engineering | 33.3 | 45.9 | +12.7 |
| BuseyBench SVG | Perception & media › Generation | 6.6 | 5.4 | −12.3 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 34.3 | 22.0 | −12.3 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Professional › Legal | 11.4 | 22.7 | +11.3 |
| FORTRESS Average Risk Score (ARS) | Safety & alignment | 22.7 | 11.9 | +10.8 |
| Legal Research Bench | Professional › Legal | 38.9 | 49.0 | +10.1 |
| SimpleBench | Knowledge & reasoning › General knowledge | 82.4 | 72.7 | −9.7 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 10.5 | 20.0 | +9.5 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 38.2 | 47.1 | +8.9 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 18.7 | 27.2 | +8.4 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 23.7 | 7.7 | −8.0 |
| Ramp SWE-Bench v2 (Bash harness) | Software › Software engineering | 47.0 | 54.4 | +7.4 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 60.9 | 68.1 | +7.2 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 48.9 | 55.6 | +6.7 |
| Professional Reasoning Benchmark - Finance | Professional › Finance, accounting & tax | 49.5 | 56.0 | +6.5 |
| Vals Legal Index | Professional › Legal | 69.8 | 75.7 | +5.9 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 30.8 | 36.0 | +5.2 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 78.7 | 83.7 | +5.1 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 44.2 | 49.2 | +5.0 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 66.8 | 71.4 | +4.6 |
| Vals Index v2.2 | Composite indexes | 56.0 | 60.5 | +4.5 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Tool & computer use › Customer service & operations | 48.8 | 53.3 | +4.5 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 63.2 | 58.8 | −4.4 |
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 6.0 | 2.0 | −4.0 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | -2.8 | 4.8 | +3.8 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 58.5 | 62.1 | +3.6 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 4.0 | -2.7 | −3.4 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 26.5 | 29.8 | +3.3 |
| RiemannBench (November 2026 rescoring) | Math | 56.8 | 53.6 | −3.2 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1585 | 1608 | +3.2 |
| Artificial Analysis Progress Index v1 | Composite indexes | 320 | 329 | +3.0 |
| GDP.xlsx | Professional › General knowledge work | 18.3 | 15.4 | −2.9 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 65.1 | 62.3 | −2.9 |
| APEX-Agents 1.1 | Professional › General knowledge work | 72.9 | 75.7 | +2.8 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 15.6 | 18.3 | +2.8 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 81.7 | 79.3 | −2.4 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 65.8 | 63.8 | −1.9 |
| APEX-1 (Mercor, mean score) | Professional › General knowledge work | 68.8 | 70.4 | +1.6 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 1.0 | -2.0 | −1.5 |
| SWE Atlas v1.1 - Overall | Software › Software engineering | 57.3 | 56.0 | −1.3 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 8.9 | 6.6 | −1.1 |
| ParseBench-2 | Perception & media › Documents & charts | 75.5 | 74.5 | −1.0 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 1.5 | 1.0 | −1.0 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 72.9 | 72.2 | −0.7 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.0 | 0.9 | +0.5 |
| Vals Finance Index | Professional › Finance, accounting & tax | 60.5 | 60.9 | +0.5 |
| GBENCH One-shot Coding GScore | Games | 56.6 | 56.4 | −0.3 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 22.8 | 22.8 | 0.0 |
Data: Artificial Analysis, Vals AI, Surge AI and 10 othersatlas.village.dev